Setting up a multi-company Paperclip deployment
Multi-company deployment is needed in two scenarios: an agency serving multiple clients from a single installation; or a holding company with multiple legal entities. Each company has an isolated AI organization, but everything runs on a single infrastructure.
Insulation architecture
Database-level isolation: A separate PostgreSQL schema per organization. Strict data separation at the RLS (Row Level Security) level. An agent from company A cannot access company B's data.
Agent namespace isolation: Agents receive context only for their organization. LLM calls are marked with an organization_id. Logs are completely separate.
Resource isolation: Separate task queues (Redis namespacing). Rate limits per organization. Billing per organization.
Network-level (optional): For high isolation requirements, separate Docker networks per organization.
Admin Layer
Super-admin panel for managing all organizations: creation, tariff settings, usage monitoring, and billing. Organization admins see only their own organization.
Development pipeline: 3–5 weeks
Weeks 1–2: Designing a multi-tenant scheme. Database isolation setup.
Weeks 3–5: Admin API and organization management panel. Billing integration. Isolation tests.
Scaling
Kubernetes allows for dynamic resource addition as the number of organizations grows. Horizontal scaling of agent workers. Per-organization resource utilization monitoring.